Job instance details

MaxCompute Studio provides two ways to view job instance details.

• Open job details in read-only mode by using a Logview URL or a local offline Logview file.

  Logview is a common way to view job details, including the status of jobs in other projects. To do so, enter a valid Logview URL or import a local offline Logview file:

  In the IntelliJ IDEA menu bar, choose MaxCompute > Open Logview. In the Open job detail by logview dialog box, paste a valid Logview URL or import a local offline Logview file.

• Double-click a job in the job list to view the details of the instance.

The job details page consists of a toolbar at the top, a basic properties pane on the left, and a details pane on the right. The details pane contains the following views:

• Execution view: Displays the overall job information as a DAG. You can view dependencies between subtasks and the detailed execution plan for each subtask.

• Analyze: Displays job run details by using the Timeline view, scatter plots, data skew charts, and long-tail charts.

• History: Shows the duration of each stage of the job run.

• Script: Displays the SQL statement submitted for the job and the parameter settings used at submission.

• Details: Displays detailed job information in a table view, including the subtask list, the worker list for each subtask, worker throughput, execution time, and status.

• Summary: Displays detailed job run information in JSON format.

• Result: Displays the job output.

• Plan: Displays job execution plan information.

The Execution view is a primary tool for daily use. It visualizes the dependencies among Fuxi Jobs, Fuxi Tasks, and Operations, and provides auxiliary tools such as Job playback, progress charts, and heat maps to help you troubleshoot issues.

The Execution view shows dependencies at the Fuxi Job, Fuxi Task, and Operation layers. By default, the Fuxi Task layer is displayed. You can click the up arrow to switch layers.

• Fuxi Job layer

  Click MaxCompute job to open the Fuxi Job layer. Each node displays the Fuxi Task name, start time, and end time.

• Fuxi Task layer

  Double-click any Fuxi Job node to enter the Fuxi Task layer.

  If there are multiple Fuxi Jobs, the Fuxi Task layer for the last Fuxi Job opens by default. This layer displays Fuxi Task dependencies, input and output tables, and partitions. After the job is complete, click Progress Chart in the toolbar to select a heat map, such as Input Heat Chart (Record), Output Heat Chart (Record), Input Heat Chart (Byte), Output Heat Chart (Byte), TaskTime Heat Chart, or InstanceTime Heat Chart. The progress chart shows the completion progress of nodes, and the heat map uses colors to indicate the hotspots of different nodes.

  A Fuxi Task node contains the following information:

  • Instance Count: Displayed in a/b/c format, where 'a', 'b', and 'c' represent the number of running, completed, and total subtask instances, respectively.

  • I/O Records: The number of input and output records at a specific point in time.

  • Percentage and orange progress bar: Indicates the run progress of the task. This percentage is calculated based on an analysis of the running subtask instances.

  • Lines between subtasks: Show the number of output records. The arrows indicate the direction of data flow.

• Operation layer

  Double-click any Fuxi Task to open the Operation layer.

  The Operation layer reveals the internal workings of a Fuxi Task. Click any node to view its complete information.

This view is mainly for SQL DML jobs. It displays the list of Fuxi Tasks and compute workers on the cluster. A job typically maps to one or more Fuxi Jobs. Each Fuxi Job is split into multiple Fuxi Tasks (stages), and each Fuxi Task contains multiple Fuxi instances (workers). You can right-click a Fuxi instance to view the standard output, standard error, and debug information for the job run.

The numbered items in the view are described as follows:

1. Fuxi Job tab.

2. Fuxi Task list.

3. Details and compute worker list for each Fuxi Task.

Click the > button to start the playback, and click it again to pause. You can also drag the progress bar manually.

The Timeline view displays detailed data about the job's distributed execution in a Gantt chart. You can adjust the display granularity to show each compute worker. The Gantt chart helps you quickly spot time bottlenecks and long-tail workers. By using various filters, you can identify the job's critical path, the workers that processed the most data, and the workers with the longest execution time.

The Analysis view displays the long-tail workers and workers with data skew for the job. It uses scatter plots and bar charts to help diagnose performance bottlenecks. From the plots and charts, you can click a node to view the details of the corresponding Fuxi instance.